Company Interview Questions · Consulting & Big 4

KPMG: Data Scientist (GenAI focus)

❓ 6 questions🏢 KPMG🆕 Recent reports

KPMG: Data Scientist (GenAI focus) Interview Experiences (2025–26). Rounds, real reported questions and model answers, compiled from public candidate reports (sources at the end). Practice mode hides every answer: say your answer first, press Show answer, then rate yourself. Use 🔁 Review again to build your revision list and 🎲 Random question for a quick drill.

★ Must-know 145📝 Assignments
About this interview

Compiled from four public candidate reports for KPMG Data Scientist roles (August 2025 – April 2026), including one GenAI-focused role. The questions are paraphrased, and the model answers are written by EasyLearnSpace.

Role Data Scientist; Data Scientist (GenAI); GenAI Consultant / Assistant Manager listings in India
Rounds Usually one main interview (plus HR); about 2 months from application in one case
Topics Imbalanced data, model evaluation, time series, Bayesian methods, price / cost optimisation, why KPMG
Observation "GenAI is a major focus"; they want people who can apply AI to client problems
Outcomes 1 offer (declined), 3 not selected

01Rounds at a glance

02Questions and model answers

Q1
How would you approach price and cost optimisation for a client? What data do you need?
  • Clarify the objective: maximise margin or revenue, or hit a volume target? What constraints apply (minimum margin, competitor parity, contracts)?
  • Data:
  • Transaction history (price, quantity, discounts).
  • Product attributes and cost (COGS).
  • Competitor prices.
  • Seasonality, promotions and channel.
  • Customer segments.
  • Inventory.
  • Method: 1. Estimate price elasticity per product or segment (log-log regression, or hierarchical Bayesian for sparse products). 2. Model demand = f(price, competitor price, season...). 3. Optimise price within the constraints (grid search or linear/non-linear programming). 4. Validate with A/B or regional tests.
  • Cost side: analyse spend by supplier and category, find outliers, and compare against should-cost models.
  • Deliver: a pricing recommendation dashboard with the expected impact and its risks.
Q2
How do you handle imbalanced data?
  • Metrics first: PR-AUC, F1, recall at a fixed precision. Never accuracy.
  • Data level: undersample the majority class, oversample (SMOTE) only on the training folds, collect more minority data.
  • Algorithm level: class weights (class_weight="balanced", scale_pos_weight), focal loss, anomaly detection for extreme rarity.
  • Decision level: tune the threshold on business costs, and calibrate probabilities.
Q3
How do you evaluate a predictive model properly?
  • A split that matches reality: time-based for forecasting and fraud; grouped by customer to avoid leakage.
  • A metric tied to the decision: RMSE or MAPE (forecasting), AUC/PR-AUC/F1 (classification), plus calibration.
  • Compare against a baseline (naïve forecast, a rule, last-period value).
  • Error analysis by segment, stability over time, and an explainability check.
  • Business evaluation: expected savings or revenue at the chosen threshold.
Q4
How do you approach a time-series forecasting problem?
  • Explore: trend, seasonality (daily, weekly, annual), holidays, structural breaks, missing data.
  • Baselines: naïve, seasonal naïve, moving average.
  • Models:
  • Statistical: ETS, ARIMA/SARIMA, Prophet.
  • ML: LightGBM with lag, rolling and calendar features, which works across many series.
  • Deep learning: N-BEATS or TFT for large panels.
  • Validation: rolling-origin (walk-forward) backtesting. Never shuffle.
  • Metrics: MAPE/sMAPE, WAPE, MASE; prediction intervals for planning.
Q5
When would you use Bayesian methods in a business problem?
  • Small or sparse data, where priors stabilise the estimates: hierarchical models for many stores or products sharing information.
  • When you need uncertainty: credible intervals for decisions (inventory, pricing risk).
  • Bayesian A/B testing: "the probability that B beats A is 96%", which is easier for business teams to read; it also allows continuous monitoring.
  • Tools: PyMC, Stan, NumPyro. Mention the trade-off: more compute, and you must justify the priors.
Q6
Why KPMG, and how would you apply GenAI for KPMG clients?
  • Why: "I want to solve varied, high-impact client problems in regulated industries, where trust and governance matter. KPMG's audit, tax and advisory work is exactly where responsible GenAI adds value."
  • Use cases to mention:
  • Contract and policy review with RAG and citations.
  • Audit-evidence summarisation.
  • Tax-query assistants.
  • Risk and compliance monitoring.
  • Finance-close automation with agents and human approval.
  • Emphasise governance: data privacy, explainability, human review and audit trails.

03How to prepare for KPMG

Sources

Created by Ajeetkumar